SourcesTaalas HC1 Custom Silicon 17000 Tokens Per SecondTaalas·high signalXBlueskyLinkedInCopy linkCanadian startup serving Llama 3.1 8B at 17,000 tok/s on model-specific ASIC. 73x H200 speed, 10x less power. TSMC N6, M raised from Fidelity.SourceSource pageTaalas↳ Follow the threadStack layer / Threat patternTrojanized pantheon-agents wheels on PyPI ship a Bun-based credential stealer, GitHub source untouchedGitHub Advisory DatabaseStack layer / ContrastGoogle Cloud Added Deferred-Execution Agent Pricing at Half the Inference Cost and Pooled Quota Across Business Apps and IDEsGoogle Cloud BlogStack layer / ContrastUSEReady's MigratorIQ Puts Five Agents on Legacy BI Migration With a Fidelity Score and a Per-Agent Token CeilingUSEReady via EIN PresswireStack layer / Update threadVS Code 1.135 shows Copilot and Claude agent sessions started in other apps, and adds a second-opinion model commandVisual Studio CodeStack layer / Threat patternTalos Sells a Deterministic Permission Kernel Around Claude and Ships 2,242 Tests as the ProofTalos (Show HN, 2026-08-28)Policy dependency / Stack layerLMSM Ports the Linux Security Modules Split to LLM Serving, Cutting HarmBench ASR 39.20% to 3.32% at 98.14% ThroughputarXiv 2608.25697Stack layer / ContrastWarp's self-improving agents run a second 'improver' skill on a schedule that rewrites the first skill from PR feedbackClaude by AnthropicStack layerQwen3.8 27B at UD-IQ3_XXS Fits Over 200k Context in 16GB VRAM, at the Cost of Half the Prompt Processing Speedr/LocalLLaMA